mv_machine_learning: tune face recognition

[Version] : 0.18.4-0
[Issue type] enhancement

Tuned face recognition accuracy by dropping flip data augmentation and
chaning decision threshold value.

Change-Id: I505628a1d64ff4460eb2760bc9cedb40fd7e65a8
Signed-off-by: Inki Dae <inki.dae@samsung.com>

diff --git a/media-vision-config.json b/media-vision-config.json
index 660c9f4cd454e6e4d381a36111602d2e187b1010..02cea2b4ba1eb0ca23f4a74b2d3f5eec8edc51dc 100644 (file)
--- a/media-vision-config.json
+++ b/media-vision-config.json
@@ -261,7 +261,7 @@
         {
             "name"  : "MV_FACE_RECOGNITION_DECISION_THRESHOLD",
             "type"  : "double",
-            "value" : -0.97
+            "value" : -0.85
         }
     ]
 }

diff --git a/mv_machine_learning/face_recognition/src/face_recognition.cpp b/mv_machine_learning/face_recognition/src/face_recognition.cpp
index d4df50b2964976b90c35865919f8cf539a8255cb..692cc05120e169d93d614b6de3756a6e84b49331 100644 (file)
--- a/mv_machine_learning/face_recognition/src/face_recognition.cpp
+++ b/mv_machine_learning/face_recognition/src/face_recognition.cpp
@@ -41,7 +41,6 @@ FaceRecognition::FaceRecognition() :
                _initialized(false), _prepared(false), _internal(), _backbone(), _face_net_info(), _training_model(), _label_manager()
 {
        _data_augments.push_back(std::make_unique<DataAugmentDefault>());
-       _data_augments.push_back(std::make_unique<DataAugmentFlip>());
        /* Add other data argument classes. */
 }